Can machine learning measure how much radio bursts blur as they travel?

Measuring how much interstellar material distorts fast radio bursts (FRBs) is currently slow and error-prone. Researchers built MT-GMDN, a deep learning system that simultaneously analyzes two different views of each burst—its dynamic spectrum and time profile—through parallel neural networks, then outputs a probability distribution for the scattering parameter rather than a single guess. Tested on 3,500 CHIME/FRB events, it achieves 94% accuracy and correctly identifies bursts with no measurable scattering 90% of the time.